Dall-E 2
Dall E 2 #
When I first heard about Dall-E, I was impressed and when I got to know that it will have a invite only beta I registered for it. It was around April when I registered and soon after forgot about it.
I got an email from openai on 16th June, it was invitation for trying Dall-E 2. I had some big expectations after seeing examples on its homepage but it blew them all away. It is something you truly grasp after trying it yourself. Since it has already been 3 months since Dall-E 2 came out, everything about it already has been written already. So, I will skip that part and move directly towards what i made with it.
Process #
There are basically three ways you can give Dall-E something to work with. First one being a simple text prompt, second one an image, then it will try to generate variations. And last, the one tried least, edit an uploaded or Dall-E generated image. By edit I mean remove a part of image and Dall-E will try to fill it for you.
oil painting of an annoyed cat wearing a crown |
a fortune-telling shiba inu reading your fate in a giant hamburger, oil painting |
Dall-E is very very good with drawing animals in any art form doing whatever, wearing whatever. It still suffers somewhat in second example, but again “reading fate in a giant hamburger” is vague to say the least.
a chocolate house with cream as clouds and chimney made from sweet straw |
But sometimes making it understand some things is a tough ordeal. For example above only 2 examples are somewhat close.
clay-coated cottage, standing at the end of the cobbled tread at night |
small house at the end of a cobbled road with a small yellowed weed choked courtyard with two chairs |
a charcoal black table with a mist filled cracked snow globe and stacks of postcards scattered all over it |
These were some of the photo-real outputs with a long description. I was ready to get surprised by even somewhat matching examples but these outputs were beyond my expectations.
a violin carved from maple bark |
And it can items that don’t exist and maybe that’s why it suffers in detailing.
Synth wave Darth Vader Chilling in his Bedroom |
It is aware about pop-culture not just items and things and that in itself is pretty incredible. But it can create stuff with it which is even more fascinating.
minimal dog |
Honestly this was the example that reallllly blew me way, it knows its a minimal picture of dog and then create other minimal pictures using same art-style. It can relate between a minimal outline of a dog and that is just mind-blowing.
Overall it is really incredible, well as expected from OpenAI. It even know how to create images with description of lens imaginary camera, extend paintings, create never realised art and much more. All it takes is some imagination, I hope which you have plenty of. If you got some requests to make regarding Dall-E, I’ll happy to generate them for you.